k-means algorithm

Nowadays, money laundering has become a serious threat to the world economy. Traditional methods of Anti Money Laundering (AML) are costly and inefficient. Recently, data mining techniques have been developed and have been considered as appropriate methods to detect money laundering activities. The purpose of this research is to detect money laundering suspicious cases which might need more detailed scrutiny using data mining algorithms with real banking transaction datasets. CRISP-DM would be used as the research methodology, the statistical population would be the banking transactions and samples would be the transactions of one of the bank branches. For this purpose, two main approaches are used. In the first approach, using the k-means algorithm, financial transactions of banking accounts are clustered. Then, using anomaly detection techniques, abnormal transactions that might be suspicious of money laundering and need to be scrutinized in more detail have been detected. In the second approach, a novel technique using Benford’s law and GANs algorithm has been introduced. It can detect financial accounts that used concocted amounts in their transactions and might be suspicious of financial fraud and money laundering. The first approach can identify accounts with outliers in their transactions with an accuracy of about 93%, and the second approach can identify suspicious accounts that do not use professional methods to hide fake figures in their transactions with an accuracy of about 60%. to recognize correctly.

Clustering as a process to understand the nature and structure of data plays an important role in organizing data in many areas of science and technology. One of the most widely used and simple algorithms for clustering is K-means. The present study was conducted to systematically reviewing research on improving K-means algorithm on data clustering. This research examines the researches conducted in this field and its role in organizing data in the range of 2010 to 2020 with a new strategy based on the shortcomings of the K-means algorithm. For this purpose, the amount of attention of researchers to eliminate any of the shortcomings of this algorithm in order to improve it in recent years has been compiled in the form of research questions. In this study, with the use of a search strategy for refining and extracting articles, 47 related sources were identified and examined. Findings showed that most researches have been done by overcoming the sensitive shortcomings to initial cluster centers to improve the K-means algorithm. Also, out of a total of 47 studies, the improved K-means algorithm has been applied in 35 studies on non-textual data and in 12 studies on textual data. Finally, the results of a review of six studies showed that the amount of data is directly related to the performance of improved K-means algorithm. In other words, this algorithm must be modified in such a way as to perform efficient and accurate clustering by applying it to different amounts of data.

In the insurance industry, customers’ systematic identification and clustering is a major concern not only for marketers but for the entire organization, for this reason, Customer segmentation helps target organizations to customize their services and prioritize products based on their profitability. This research is an applied, descriptive and quantitative study aiming to cluster customers by using k-means clustering. The data were collected from 800 customers of Pasargad insurance company in the city of Shiraz using the random sampling technique. The data on length, recency, frequency and monetary issues were collected by considering research ethics principles. Customers were clustered into four groups including key, prodigal, intermittent and uncertain by using the K-means method. Eventually, the customers’ lifetime value was determined Clustering has been carried out in four categories, including key clients whose contribution to a sample of 800 insurance customers is 24.2%. This group of customers has high financial value characteristics and high purchase frequencies. They are ranked first in terms of lifetime value. Based on the findings, the indicator of the volume of financial exchange is an index that graduates the other indices placing a client in the position of key a customer. Prodigal customers featured with high financial characteristics, low shopping frequencies and a 25.8% share of insurance customers are in the second category and ranked second in terms of lifetime value. The third group of customers, having a share of 33.4% of the insurance customers, low purchasing value characteristics and high purchase frequency, are frequent customers who are in the third rank of life value. The last group of customers is uncertain ones who account for a significant 16.6% share of customers. They have monetary value characteristics and low purchasing frequency and are ranked last in terms of lifetime value. They are among the customers who have no significant trade volumes and the lowest value of the purchasing iteration index, regardless of the time indicators associated with these customers. This puts them in the cluster of uncertain customers with a 16.6% share in the selected statistical sample. This is because they have different and irregular financial behaviors during a certain period. So, it may not be profitable to give them services.   Determining the share and importance of customer groups based on customer lifetime value is one of the results of this study. While keeping prodigal customers, it is recommended to managers and marketing planners of the insurance industry to pay special attention to key and intermittent customers. From a managerial perspective, customer segmentation is a very important issue in the insurance industry. It can be a subject for studies and applied planning in every sector. Also, the specialization of insurance industry services in proportion to the customers' lifetime value, expectations and preferences based on scientific segmentation and customer data is one of the managerial recommendations. Another aspect that can be suggested to the managers of the insurance industry based on the results of this study is paying attention to the characteristics of customers in each cluster. Among these four groups, the cluster of key customers has a significant volume of transactions and length of the period of communication and repetition of insurance transactions. It also requires insurance companies to pay special attention to these customers. Next to this group are prodigal customers who have mostly low repetition of their insurance transactions, while the volume of turnover of this group is significant for the insurance industry. The importance of this group increases when these people have the lowest share in the overhead costs of insurance services for insurance companies, and, at the same time, their premiums are relatively higher than other groups. This makes managers pay more attention to this group. However, due to the low contact of these people with the employees of insurance companies, it is possible that they will receive less attention in relational marketing issues and promotional measures of this group. Accordingly, it is necessary for the managers of the insurance company to recognize generous customers and make special plans for them, especially in relationship marketing. In addition, given that a good number of the insurance company customers are uncertain clients, special planning is necessary to maintain and increase their loyalty. Another group identified in this study is that of the intermittent customers. This group of customers receive a relatively large amount of insurance services, while the premiums received from this group are not significant compared to the other groups. Identifying this type of customers and defining ways to retain them while reducing referrals to this group of customers is essential.

In today's competitive world, applying new techniques to business development has a great impact. The restaurant industry is no exception. Therefore, in this research, using new methods of knowledge discovery and data mining, customer data of chain restaurant is investigated. The purpose of this study was to explore customer behavior patterns using data mining methods.In this study, one million and five hundred thousand customer records were reviewed in five branches of a chain restaurant and two stages of clustering modeling using RFM method and then classification modeling were performed on the data and the behavior rules chain restaurant customers were extracted. The results of this study have helped to identify the loyal and profitable customers of the chain restaurant which has led to the improvement of the profitability of the chain restaurant. One of the innovations of this research has been the communication between clustering and classification results.

The knowledge-based economy is the newest pattern of production in the current era. So far, this pattern has resulted in unique achievements for a wide range of countries. This study aims to classify the provinces of Iran in terms of Knowledge-based economy. The classification of provinces based on their similarity in achieving the knowledge-based production pattern is the first step for correct and realistic planning. The same version cannot be used for different provinces. The regional knowledge-based economy index is defined in three dimensions: education, innovation, and information and communication technology, based on 15 sub-indices. The classification is based on the clustering technique, which is one of the branches of unsupervised learning. To do this, k-means and fuzzy c-means algorithms are used simultaneously to compare their results. The optimal number of clusters is calculated through the Silhouette coefficient. This coefficient also indicates the accuracy of the clustering results. Clustering based on the fuzzy c-means algorithm in 6-cluster case with a Silhouette coefficient of 0.77 is the most appropriate classification for research purposes. The results show that there is a clear discrepancy between different provinces in the context of knowledge-based economy. Tehran and Alborz are in separate clusters and are among the leading classes compared to others, while more than half of the provinces belong to backward cluster.

In recent years, a new issue called "multi-criteria clustering" has emerged that aims at grouping alternatives into homogeneous classes called clusters according to different evaluation criteria. Following the related studies in literature, by combining K-means algorithm and PROMETHEE technique, this paper aims to present a new multi-criteria clustering method. The parameters of the problem are the cluster separator profiles which genetic algorithm (GA) is used to optimize them. In the modeling process in each stage of updating responses, alternatives allocate to the nearest cluster according to the distance of their pure flow of privileges from the profiles. The mutation operator is only applied when the chromosomes’ similarity level in each population reaches to a certain level which this intelligence reduces the computation time. Finally, by simulating the proposed algorithm and some well-known clustering algorithms based on the several financial databases the efficiency of the algorithm compared to other algorithms. The results show the algorithm, in addition to determine the optimal number of clusters in comparison to other algorithms, also provides better results.

With the increasing growth of Web-based resources and articles, the use of quick and inexpensive ways to access the texts is important from the vast collection of these documents. The main objective of this research is to cluster the base of Iranian Research Institute for information Science and Technology (IranDoc) based on text mining techniques. So that the articles are ivided into several clusters so that the articles of the different clusters have the maximum possible difference and the articles in each cluster have the most similarity. Articles on information technology related fields were selected. For this purpose, first all the keywords of information technology fields were selected based on their frequencies in base articles and then the articles of each keyword were extracted from the Iran Doc database. Then, using the notepad ++ software, the dataset was created. In this research, clustering of k_means algorithm and Euclidean distance function criterion were used to measure the similarity of clusters. Then the results of the clustering were analyzed to find the similarity and pattern among the papers. The pattern showed that the greatest similarity is found between articles in the two data mining clusters and the neural network with an Euclidean distance of 1.365, and the least similarity between the two cluster articles is optimization and image processing with a distance of 1.387. Research knowledge, clustering of articles related to the highest and the least degree of similarity with each other, finding a new pattern for quick and easy access to similar articles, and discovering hidden relationships among different subjects. This knowledge helps researchers to access topic-related articles related to specialization Identify themselves and the subject of the study in a more desirable way.

The aim of this paper is to present a hybrid model to evaluate performance of loan portfolio of banking system regarding loan repayment status and to forecast credit status of loan applicants. At first stage, we have taken credit granting management approach in order to cluster and rank 100,224 loans granted by Karafarini Omid Fund. All the data on the loans granted to clients was extracted from core banking software of the Fund. Because of having access to this valuable and valid dataset, qualitative data collection methods are not used. In the first section of paper, a type of robust principal component analysis (ROBPCA) was utilized to classify the clients. Then, the eigenvector derived from ROBPCA was used as input to a two-step K-means clustering algorithm. Then, to propose a model to forecast credit status of applicants prior to granting loans, support vector machine (SVM) and artificial genetic neural networks were used. The results obtained from the applicants’ credit status forecasting showed that the model based on the artificial genetic neural networks with the mean-square error of 0.23 and %78 coefficient of determination leads to more accurate forecasting than support vector machine. Therefore, the proposed model for forecasting the applicants’ credit status can predict their performance with relative accurately. A new method in the form of data mining software provides credit institutions with the possibility of predicting applicants’ credit regarding loan repayments.

This paper wants to extract online media branding. The method of this research is datamining and in terms of it’s applicable. The research questionnaire was extracted from previous studies and included content, audience, media context and environment dimensions. The researcher-made questionnaire, after calculating its validity and reliability, was provided to audiences who visited the Varzesh3 more than 10 times a month. The statistical population of the study consisted of 800 thousand people per day. Using a systematic random sampling, 100 questionnaires were included in the sample size (8 thousand people). About 50 percent of them 4056 responded to the questionnaire. The response of the audience is extracted using the K-means algorithm processing and branding model. The results suggest that in order to increase the life of the audience, the professional components of the media, including reference, bias, newsworthiness and the speed of the publication of the news as "brand perceived quality" should be considered. The loyal behavior of the audience is also due to demographic factors. The clustering of a loyal audience with a center of age indicates that the largest cluster with 44% of men with Bachelor education and ranges from 26 to 35 old. “Live scores” and “league standing” are the varzesh3 of brand identity. It is suggested that this media pay more attention to the creation of Brand Identity for women and audiences under the age of 15.

Traffic analysis zones represent locations of trip generation and attraction. These zones are widely used in many urban transportation planning activities. Proper design of the mentioned zones is very important, because they cause considerable effects on the results of different transportation modeling and analyses. In this paper an efficient method is proposed to design these zones using geospatial information system. By using this method, various criteria can be considered to determine homogeneous traffic analysis zones whereas the number and size of the zones can be controlled as well. In the proposed method, the k-means technique is used for clustering basic areal units to homogeneous traffic analysis zones. To prevent overlaps and gaps between the zones and creation of the zones with too large or small sizes, some constraints are added to the k-means technique. However, definition of homogeneous zones without considering their geometric compactness results in generation of zones with complex shapes. For solving this problem, a geometric algorithm is developed to create continuous and compact zones. Efficiency of the proposed approach has been tested through to redesign the existing traffic analysis zones of Tehran city and its comparison with the existing traffic analysis zones. Results showed that the redesigned traffic analysis zones with the proposed method have been more homogeneous than the existing zones. Also, by using the shape compactness algorithm, compactness of the resulting zones increases whereas their homogeneity decreases. However the redesigned zones are more homogeneous than the existing ones.